Recommendations

Shopping recommendations are a great mechanism to drive revenue for retailers and to provide guided exploration of product to consumers. I buy a lot of stuff from Amazon, and one of the ways I explore what they have to offer and make additional purchases is via their recommendations section. It's generally quite good, but it can also be a bit of a blunt instrument. Here is a case in point:

Because I purchased an anthology collection of graphic novels (that's comics to the uninitiated/uninterested) about Transformers, yes robots that turn into cars and planes, I have been recommended a memory card. Not just any memory card, but one that appears to be a surprisingly over priced single-system media card for a system (a PS Vita) that I do not own, and for which Amazon has no indication that I may own one. This is a lost opportunity and diminishes the real and perceived value of the recommendation service. It is difficult, though of course not impossible, to imagine that there is a significant statistical trend behind this recommendation.

Here are some helpful handy tips for building a better recommendation system and user experience:

• Focus on quality. A bad suggestion is probably worse than no suggestion. You devalue your service and lower the trust and positive perception of your service, and possibly your expertise and good standing. Similarly, recommendation offerings based purely on revenue (e.g. suggesting accessories with the highest profit margin) may increase your revenue, but may also drive customers away.
• Make sure your recommendations are based on a significant sample and effect size, if you do not you're offering poor quality recommendations. If your recommendations are provided by human experts, then make sure they are experts and are providing useful recommendations.
• Recommendations should be targeted, relevant where possible, and within the buying/consideration/interest scope of the user.
• Factor in customer ratings and feedback into your recommendation system; it provides reassurance and supports social confirmation.
• In general consider avoiding recommending low rated product to consumers. It won't do your brand image any good.
• Allow feedback on recommendations (as Amazon does, as shown in the screenshot) and apply it widely; your customers are telling you what they like and don't like, so stop offering them things that are similar to lots of other things that they don't like.
• Support customisation for recommendations, for example allow users to add additional weighting to some categories or to exclude certain types of product and media. For example, I have an XBox 360 but not a Playstation and no intention to buy one. Companies like Amazon would be better served in providing game recommendations to me in a relevant format (i.e. the XBox version rather than the Playstation version), or to provide recommendations for different items altogether.

A good recommendation system is a win-win, it's good for the retailer and the consumer.